Skip to content

08_sensitivity_analysis

Allows the testing of many parameter combinations, running three detection steps mask_vi_from_dataframe, train_model_from_dataframe and dieback_detection_from_dataframe using default parameters as well as user defined parameter combinations.

A 'test_info.csv' is written in 'testing_directory', where each test_id is associated with the value of all parameters used in the iteration.

sensitivity_analysis

INPUT PARAMETERS

  • testing_directory : str : Directory where the results will be exported.
  • reflectance_path : str : Path of the csv file with extracted reflectance.
  • args_to_test : dict or str : Either a dict where each key is any argument to functions mask_vi_from_dataframe, train_model_from_dataframe or dieback_detection_from_dataframe, and the values are list of values to test. All combinations will be tested.If a str is given, it is interpreted as the path to a text file where each line begins parameter name, then each value is separated with a space. See an example here
  • cloudiness_path : str, optional : Path of a csv with the columns 'area_name','Date' and 'cloudiness' used to filter acquisitions, can be calculated by the extract_cloudiness function. Not used if None.
  • update_masked_vi : bool, optional : If True, updates the csv containing the vegetation for each acquisition with the columns 'period_id', 'state', 'predicted_vi', 'diff_vi' and 'anomaly'. Doesn't apply if 'overwrite = True'. The default is False.
  • name_column : str, optional : The default is "id".

OUPUT

Three csv files are written in testing_directory :

  • test_info.csv, which contains the fordead parameter used in each test, with the following columns :
    • test_id : A unique ID for the iteration
    • lim_perc_cloud, ..., threshold_anomaly : The value of each fordead parameter in a different column
  • merged_pixel_info.csv, which contains information at the pixel level, with the following columns :
    • epsg : The CRS of the Sentinel-2 tile from which data was extracted
    • area_name : The name of the Sentinel-2 tile from which data was extracted
    • an ID column corresponding to the name_column parameter
    • id_pixel : The ID of the pixel
    • last_training_date : The last date used for training
    • coeff1, coeff2, ..., coeff5 : Value of the corresponding coefficient of the harmonic model
  • merged_periods.csv : which contains the merged detected periods information for all iterations, with the following columns:
    • area_name : The name of the Sentinel-2 tile from which data was extracted
    • an ID column corresponding to the name_column parameter
    • id_pixel : The ID of the pixel
    • period_id :
    • state : The period state, which can be :
      • Training : Period used in training the harmonic model
      • Healthy : Period detected as healthy, with no stress, dieback or bare ground detected
      • Stress : Period beginning with 3 successive anomalies, ending with the last anomaly before three successive non-anomalies of the beginning of a 'Healthy' period.
      • Dieback : Period beginning with 3 successive anomalies, ending with the last available acquisition, or the beggining of a Bare ground period.
      • Invalid : The pixel is invalid, there were not enough valid acquisitions to compute a harmonic model
    • first_date : The first date of the period
    • last_date : The first date of the period
    • anomaly_intensity : Depending on stress_index_mode, either a mean or a weighted mean of the difference between the calculated vegetation indices and their predicted value for the period. The weight is the number of the date within that period (1+2+3+...+ nb_dates). It is only calculated for 'Healthy', 'Stress' and 'Dieback' periods

Running this step

Using a script

from fordead.validation.sensitivity_analysis import sensitivity_analysis

sensitivity_analysis(testing_directory = <export_dir>,
        reflectance_path = <reflectance_csv_path>,
        name_column = 'id',
        update_masked_vi = False, #Computes faster
        args_to_test = args_to_test)

From the command invite

This step can also be ran from the command prompt. 

fordead sensitivity_analysis [OPTIONS]

See detailed documentation for the command line prompt here